Let $A$ be a symmetric multilinear form on $\left(\mathbb{R}^d\right)^{\otimes n}\times \left(\mathbb{R}^d\right)^{\otimes n}$ and consider the random variable: \begin{align*} X=A(g_1,\ldots,g_n,g_1,\ldots,g_n) \end{align*} where $g_i\in \mathbb{R}^d$ are independent random vectors with i.i.d. random Gaussian entries of mean $0$ and variance $1$. Is there a closed form expression for $\mathbb{E}(X^2)$? In the case $n=1$ I know that $\mathbb{E}(X^2)=\|A\|_{F}^2$, where $\|\cdot\|_{F}$ is the Frobenius norm of the underlying norm. Does that generalize in any way for $n>1$? Thanks for any help!


1 Answer 1


For $n=1$, $$EX^2=(\text{tr}\,A)^2+2\|A\|_F^2, \tag{1} $$ not $\|A\|_F^2$.

Indeed, writing $A=(a_{ij})$ and $g_1=(Z_1,\dots,Z_d)$ with iid standard normal $Z_i$'s, we have $X=\sum_{i,j}a_{ij}Z_iZ_j$ and hence $$EX^2=\sum_{i,j,k,l}a_{ij}a_{kl}\,EZ_iZ_jZ_kZ_l. $$ Next, $EZ_iZ_jZ_kZ_l\ne0$ only if (i) $i=j\ne k=l$ (in which case $EZ_iZ_jZ_kZ_l=1$) or (ii) $i=k\ne j=l$ (in which case $EZ_iZ_jZ_kZ_l=1$) or (iii) $i=l\ne j=k$ (in which case $EZ_iZ_jZ_kZ_l=1$) or (iv) $i=j=k=l$ (in which case $EZ_iZ_jZ_kZ_l=3$). So, taking the symmetry of $A$ into account, we have $$EX^2=\sum_{i\ne k}a_{ii}a_{kk}+\sum_{i\ne j}a_{ij}^2+\sum_{i\ne j}a_{ij}a_{ji}+3\sum_i a_{ii}^2 \\ =\Big(\sum_i a_{ii}\Big)^2+2\sum_{i,j}a_{ij}^2, $$ so that (1) follows.

For $n>1$, $EX^2$ will be a big sum over all partitions of the set $\{1,\dots,4n\}$ into subsets of even cardinalities (cf. cases (i)--(iv) above).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.